default iterator hack for blockwise copy (#47)

58a80570 · Chao Liu · GitHub · 1c1b56fe · 58a80570
Unverified Commit 58a80570 authored Jul 16, 2021 by Chao Liu Committed by GitHub Jul 16, 2021
Show whitespace changes
Inline Side-by-side

Showing with 10 additions and 0 deletions

composable_kernel/include/tensor_operation/blockwise_dynamic_tensor_slice_transfer.hpp ...sor_operation/blockwise_dynamic_tensor_slice_transfer.hpp +10 -0

No files found.
--- a/composable_kernel/include/tensor_operation/blockwise_dynamic_tensor_slice_transfer.hpp
+++ b/composable_kernel/include/tensor_operation/blockwise_dynamic_tensor_slice_transfer.hpp
@@ -89,6 +89,16 @@ struct BlockwiseDynamicTensorSliceTransfer_v4
        }
    }

+    template <typename SrcBuffer>
+    __device__ void RunRead(const SrcDesc& src_desc, const SrcBuffer& src_buf)
+    {
+        if(BlockSize == thread_cluster_desc_.GetElementSize() or
+           get_thread_local_1d_id() < thread_cluster_desc_.GetElementSize())
+        {
+            threadwise_transfer_.RunRead(src_desc, src_buf);
+        }
+    }
+
    template <typename DstBuffer>
    __device__ void RunWrite(const DstDesc& dst_desc, DstBuffer& dst_buf)
    {